 Good morning. Welcome to genomic medicine 15. I'm Terry Manolio. I'm told we need to stand here so we can advance the slides. So I'm co-chairing with my colleague Rex Chisholm who is standing in for Gail Jarvik who is ill. We have a couple of people felled by our favorite virus. And so Bruce Korf will be presenting remotely and a few other people as well. And we appreciate their attending even through the throws of various illnesses and that should do it. So Rex, do you want to? Sure. As Terry said, I'm standing in for Gail. I'm sure I'm a poor substitute for Gail and wish her to feel better soon. So first of all, lest they get forgotten at the end. We wanted to make sure that we recognize all the people that were important and instrumental in creating this meeting and getting it to be the success. I'm sure it will be. The planning group consisted of Jonathan Berg, Gail Jarvik, Bruce Korf and George Mensa. Thank you very much to all of them. And the organizers, Jenna Cohen, Alvaro Encinas, Brandon Michael John, Genevieve Nerula, Markuil Neurocker, Jerry Somani and Meredith Weaver. Thank you to all of you for the logistics that will also make this a successful meeting for all of us. So the goal for the meeting, and just to remind everybody, this is Genomic Medicine 15. So there were 14 Genomic Medicine meetings that have preceded this. And I think it's important to just note, and Terry will probably amplify this, but that these Genomic Medicine meetings have been, I think, very important for our community. They've led to important initiatives on the part of NHGRI as well as, I think, really exciting new science that's been developed as a consequence of this. And the reach has gone far, far beyond, I think, our community. So our goal for this meeting is to talk about the topic of genomics and population screening. And so we want to discuss the current state of population genomic screening in the U.S., as well as the barriers and opportunities to expanded population screening, the impact on clinical practice and outcomes, and various genomic screening technologies and costs, and evidence gaps that may inform future directions. So we have a few objectives for the meeting. First of all, as I already alluded to, we want to review the current state of genomic population screening. We've got some real experts here to help us with that. To examine obstacles and opportunities for expanded screening and available evidence of impact of screening on outcomes and cost. To identify research directions to inform expanded screening as appropriate. Think about variants and conditions that should be screened for. Think about populations to be screened. What is the role of community engagement in this process? And providers to order screening and managing the results. How does that work? So the meeting will address conditions which there are well-established professional guidelines for action, screenings for adults aged 18 and plus as a starting point. We're not going to focus on pediatric screening. Research directions for the entire genomics field of which NHGRI is only a small part, but obviously a mighty part. But to think about this more broadly in other conditions, NHGRI tends to be disease agnostic, but there are obviously going to be diseases, conditions where population screening is valuable. And then we would ask the presenters to identify results to be provided and interventions to be recommended when proposing conditions. And that might lead to us thinking about future studies that need to be done. What we will address is evolving an appropriate role of geneticists and genetic counselors in population screening. The potential role of telehealth and AI. That might be an interesting discussion. Importance of keeping approaches and guidelines simple and understandable. The risks of and dealing with false positives, something we all think about regularly. And research needed to make health care system ready to handle population screening. How close are we or are we long way off on that one? There are a few things that we're not going to focus on. We're not going to address newborn screening. It has its own standing committee and expert bodies. And other than something as a model and using lessons learned from a newborn screening, we're really not going to talk too much about newborn screening. We're not addressing the screening of children or parent guardian consent. We really want to focus on adults in the meeting in the next two days. We're not debating appropriateness of nor evidence behind potential interventions once a screening for condition is found. So we're going to assume that there are professional guidelines established by others that will address those. And we're not proposing funding opportunities or mechanisms. Let's focus on where the gaps in knowledge are and how we might go about filling them through research studies. So a brief note on the structure of the meeting and its aftermath. So we've built in ample time for questions and discussions, and many of the sessions have panels. We want to allow for at least one or two clarifying questions from each of our presenters, but we want to save most of the discussion for a discussion period that's been built in. The moderators will redirect from topics, the meeting will not, will redirect from topics the meeting that will not address here. We've asked the moderators to save a few minutes at the end of each session to list a few key points that have been made during that session. In-person attendees should raise hands to speak. Online attendees can put questions in the Q&A, we guarantee we'll see them or address them during the meeting, but we will try to follow up afterwards if we don't make them, make it to them in the meeting. A meeting summary aimed, the goal is to post a meeting summary in four to six weeks. That's traditional after genomic medicine meetings and generally adhered to. And if a warranted, presenters and moderators will jointly prepare a white paper for journal submission, and some of those papers have been fairly influential in the past as well. So with that, if there's no burning questions, I will turn it back to Terry. You do a wonderful Gale impression, so thank you very much. So I really appreciate you stepping in. So I just wanted to recognize and maybe give a little bit of background on these meetings where I've saluted to, they're really critical in terms of the division of genomic medicine and NHGRI planning, future studies in this field. We have an advisory group, the Genomic Medicine Working Group, which is a subgroup of our advisory council. These are the members, and they are almost all here unless they are filled by illness. And then several of us at NHGRI who help them out. The charge to the group is to assist in advising us on research needed to evaluate and move genomics into routine medical practice, reviewing current progress. Much of the goals of this meeting, as you said, now focused on genomic screening, identifying publicized key advances. That's our notable accomplishments website, which we can tell you about at some point. We don't need to talk about it here. Planning genomic medicine meetings on timely themes. And we identified population screening as a timely theme, not necessarily that it's something that would, you know, should begin tomorrow, but the research needed to determine what should be done and when probably is needed tomorrow. And then we also facilitate collaborations, coordination, et cetera. This is sort of a collage of the first 12 that we held. It's all that would fit on this slide. And you can see that there are a variety of topics. Again, as things come up and are nominated to the group, and you feel free to nominate future topics for us, we had actually had two others in addition to the 12 that were shown there. The most recent one was August of this past year, so they're roughly annually. And then this one is on population screening and genomics. So just to give you a feel for some of the products that have come out of these meetings. From the first, we had, after an initial sort of summit, we had a separate meeting that we called ClinAction, from which the clinical genome resource grew. So that was specifically designed to address variant curation and how that could be done in a more group-sourced way. From that has come the GenCC Consortium, which is an international effort, as is ClinGen, and have both been quite successful. We added pharmacogenomics to emerge shortly after that first meeting. The Ignite Consortium, our network, came out of that, looking at implementation of genomic medicine, which has grown into another phase, which I'll show in a sec. I'll skip over some of these, because we can't fit them all on one slide. The fourth meeting was on training, and it led to the Inner Society Coordinating Committee for Practitioner Health and Genomics, which is a very active group. The sixth led to the Global Genomic Medicine Consortium, and from that the International 100,000 Cohorts Consortium, both of which are active groups and engaging groups all around the world. I'd gone with sort of the major, even numbers, from the eighth meeting, which was sort of an overview of where our programs currently stood. We recognized we needed to develop some training efforts, particularly modules, to help to train clinicians in this work. From the ninth, we had a program announcement on variance functions and disease, which also brought together the basic science and clinical communities. The tenth on pharmacogenetics led to ADOPT-PGX, which is part of the second phase of the Ignite Consortium, testing pharmacogenetic interventions for three treatment scenarios. In our eleventh unemploymentation, we followed that up with a group of employers and exploring the role of employers in the possibility of genomic screening of their participants or their employees. The twelfth meeting on genomic risk prediction led to the next iteration of the Emerge Network, which is focusing on genomic risk assessment, as well as the population risk methods in diverse populations. Sorry, polygenic risk in diverse populations, that consortium, the primed consortium, which is actively pursuing that kind of research and methods. Our informatics meeting led to solicitation on patient-centered informatics, which we've received a few applications, not a lot, so keep that one in mind. Our fourteenth meeting led to a solicitation that was actually the applications were due yesterday, for genomic learning health systems, so sorry if you missed that deadline. Yes, I know, I know, what could I say? And then we brought a concept, which is sort of the first stage in bringing forward a program. We don't know that we're necessarily bringing forward a program, but it has been approved by our council to at least be considered for moving forward, and that's on electronic consults or sort of distant consults provided to provider efforts. So a lot coming out of these meetings, and we expect, we'll have to squeeze in the fifteenth meeting after we have this one. So with that, I will stop, and I'm going to pull up the next slide and turn it over to the moderator, so I think you can moderate from your seats, but I think the presenter, unless you have to come up here to advance your slides, and Eric and Mark, take it away. I am. Well, thank you, Terry. Good morning, everyone. Thank you for coming here today and tomorrow. For those of you, I'm looking around the table. I've seen a lot of you recently, like downtown D.C. last week. Thank you for enduring two weeks in a row of D.C. meetings. You can blame us for this one. Obviously, ASHG chose their location. On the other hand, seeing the double-header nature of the ASHG meeting and this genomic medicine fifteenth meeting, I did arrange for really good weather both weeks. And if you believe I have any influence on the weather around here, I'm really impressed. Any case, I'm delighted to see all of you, and I am glad there's good weather, because the falls around here can be quite nice, as you've seen. Thanks to the organizers to put this together is a tremendous amount of work, but the feedback from these meetings are always extremely positive. The productivity, as you saw with Terry, speaks for itself. Things really do come out of these genomic medicine meetings, and we always learn a lot, and it's really very valuable to NHGRI. I also want to put in a special thanks to my office communications crew who are over there with headsets on and doing various things. The other aspect of these meetings from the beginning, I think, was I guess we used to video capture it and put it up. I mean, now with Zoom, we do it live and we video capture it. The outreach of these meetings, in part because of my terrific communications crew that captures this in a very high-quality way and will get it up, and so it lives on. Besides live, we will have really good video of this that a lot of people around the world really do watch. We are all very fortunate to be here in this room, but not everybody's as fortunate. First of all, we can't have huge meetings, but more importantly, there's some people just practically can't travel here. So I really appreciate and when we really take very seriously the ability to get the information out of these meetings through our communications group. So I thank all of them for their hard work leading up to the meeting now and post-production. So Mark and I are kicking this off as the moderators. We have two speakers. We're not going to give any bio because the bios are in the books, and so if you want to read more, first of all, everybody in the room probably knows less, and Mike, but if you want to read more, I know less's bio is on page nine of the electronic PDF. I would just say in introducing less a couple of things. First of all, he and I have grown up together professionally in some ways. We both came into the intramural program at NHGRI essentially from the beginning. He arrived 30 years ago because we're celebrating our 30th anniversary of NHGRI's intramural research program. I arrived one year later, even though I committed to come the year before, and we were tenure track investigators together and then tenured and then have taken on various leadership roles, and less is currently the director and the founding director of our intramural center for precision health research. And the other thing that could say about less, which says something about the wise ideas of the organizing committee is that if you want to get discussion going, you get someone like less to give an early talk because he doesn't pull punches. He's going to tell us what he really thinks, and that will always stimulate conversation and rigorous discussion. So with less, with that, let's take it over. Is that the bowl in the China shop introduction? In an incredibly loving way. Oh, thank you. Thank you. Thanks. Okay. All right. Great to be here. I've attended a couple of these meetings and they were awesome. So I'm eager to participate with you here today. So I want to talk a little bit about the reverend here. And I think that we have to think about the reverend because what we have are genomes. We're genome people mostly. And what we need to do is convert a genome into health. That's all we have to do. And the challenge here is that little blue arrow includes, pun intended, a myriad of considerations and nuances and complexities. But that's what we have to do. And we have to make that arrow work efficiently and effectively to improve health because that is our product. Now I'm going to revive a bit of my rant from my ASHG presidential address and tell you there is nothing new about this. Genomics, genetics is just a technology. It's just a test. And it obeys, it has to obey all the laws that apply to every other test we use in the medical center. And so it has sensitivity. It has specificity. It has PPV and it has NPV. And that's the way it works. And it is no different than a hemoglobin or a sodium in that respect. And so the PPV of the test, the positive predictive, which is of course that thing that we are going to operate on is a patient with a risk of a disease. We want the genome to tell us who those people are. And that result depends primarily on the testing scenario which the topic of the seminar is about changing, the topic of this meeting, sorry. So as geneticists I have to say we're a little bit feeble in this regard because we have been spoiled by decades of practice where people come to us in a clinic with typically an enormously high prior probability of disease. These are loaded families, people who actually in many cases patients who understand the disease as well or better than we do because they've lived with it for generations and decades. And they have this context and they have an enormous high prior probability of disease. And the irony is when we do genomic testing or genetic testing in that context we actually don't change the probability of them having that disease very much by a positive result. So we're not moving them very much from a prior probability of disease to a posterior probability of disease. Now if we then all of a sudden change and shift the game to a population based game then the prior probability plummets and it's a completely different business. So this is old. This is old. This is probably a woodcut of the Reverend Thomas base. It's probably a woodcut because the historians can't agree if it was a post following his death made up a fantasy of what he looked like but they think it's him. And this in a more modern context is what the Reverend said. And it's an equation. I'm very sorry. It's math. Lots of people don't like math and in fact I think in college there's a bit of a bifurcation. People who like math and science go into physics and chemistry and those kinds of engineering and people who like science but don't like math went into biology. But here we are. We're biologists, we're geneticists and it's completely mathematical. So this equation governs not only how we interpret and classify and think about genetic diagnosis but of course again it's all diagnoses. So what's the Reverend's equation saying? There's two terms, two forms of terms in there. This notation P parentheses A is the probability of some event A. Then the second notation, the probability of A bar B is the probability that A is true if B is true. That's a conditional probability. So what the Reverend is saying here is that something A you want to know if it's true. The probability that it's true given some piece of evidence is equal to the probability of the evidence if A is true. If the thing is true. Times the probability of A, that is how likely that thing is in the first place divided by the probability B because the more common the evidence is the less likely it is that your assertion is correct. And so some people refer to this Bayes rule as the law of inverse probability because of this term because it depends enormously on the probability of B being true if A is true. All right. And so how do we then think about screening? Screening has to obey Bayes' rule. It's probabilistic. The analytic validity of the genomic test result is a probability. The clinical validity is a probability. This thing that we now call the pathogenicity of variant. That's a probabilistic measure of the validity of the relationship between the variant and some condition. This term we've coined called the clinical molecular diagnosis which is if you take the variant and the patient's presentation into consideration how likely is it that they have that underlying condition which is different from how likely it is that they have a manifestation of that condition. I'm going to talk a little bit more about that. That is penetrance and that is a separate probability from the likelihood, the probability of the clinical molecular diagnosis. And then of course, expressivity is a probability that's a qualitative distribution of the nature, not the quantity but the nature of the phenotype. And so what we need is a probabilistic model of genetic diagnosis in a screening context that takes all of those things into considerations. And we just have to admit that most people don't like probability, right? I often say when I travel there's only two kinds of people who sit next to me. The people who are sure they're going to get to their destination safe and sound. And then the white knuckles, for sure they're going to die. People don't like intermediate probabilities and there's a great story on how many of you have read The Signal in the Noise by Nate Cohn, a wonderful book. I'm sorry, Nate Silver. And he talks about the National Weather Service and they used to issue precipitation warning, precipitation probabilities and they found that the customers of the website were the least satisfied when the probability of precipitation was 50% because they didn't know what to do. Should they go to the scene of the movie or should they go on a picnic? And so what was their solution? They would randomly change it to 40 or 60% and people were happier. I'm not advocating for that but I think it's a very telling comment about people don't like probability. Okay, so what I want to frame here is a notion that all genetic disease is a susceptibility to an abnormal phenotype. We tend to think of only things like cancer susceptibility syndromes as being disease susceptibility genetic diseases and let's say birth defects as deterministic. But I think we have to consider all of them probabilistic and realize that the penetrance function goes very, very high, essentially 1.0 for some variance. And a key distinction here is that we have to recognize that having the disease but no manifestations of the disease is non-penetrance and that not having the disease is not the same as having the disease and being non-penetrant. It's a critical distinction and I think a lot of people miss that. And of course we know in variant classification nearly all variants are not certain to cause disease. They have a probability of pathogenicity that is less than 100%. So if they are not certain to be causative then you have a Bayesian problem. Which is that it depends, the diagnosis of the patient depends on their prior probability of disease because there's a possibility that the variant is in fact not pathogenic. There's some chance of that. So harboring the variant means, I'm sorry, for a few variants they essentially do have a probability of pathogenicity of 100%. There's a couple variants, probably Delta F508, a few of the HBOC variants where the evidence is so overwhelming for all practical purposes we can say the pathogenicity of that variant is 100%. In that exceptional case, then Bayes law does not apply. If you have the genotype, you have the disease, period. Whether you have a manifestation, remember, is a second consideration. That is a probabilistic function. But they have that susceptibility. And that's why we have to think of diseases as susceptibilities. So I think what we need to start doing is thinking about a stepwise approach. Which is to first, what we're doing in ClinGen is assessing the probability of the pathogenicity of the variant. And what the reverend would say we're doing is measuring the probability of pathogenicity given the evidence about a variant. And that's the predictors about the variant, historical data on the variant, et cetera. Ideally not the phenotype of the patient. And then we do a laboratory and a clinical interpretation of the person who harbors that variant. I.e., the reverend would say the probability that they have the disorder given the phenotype they have because they have the variant. And that's what we're calling a clinical molecular diagnosis. And again, harboring a variant, if the pathogenicity is less than 100%, is not the same as having the disorder. It's probabilistic. And for those who have the disorder, that is they have a clinical molecular diagnosis, but they do not have an apparent phenotype, then you have to assess penetrance. And the penetrance function is the probability of a phenotype given that they have the disorder. It makes no sense to talk about penetrance if you don't even know that the person has the disorder. So graphically, again, people don't like equations. Here's sort of what the math looks like. I published this paper a couple of years back in genome medicine, and I use a hypothetical example of diagnostic testing versus population screening for Marfan syndrome, one of the genes on the ACMG list. And then it's in a diagnostic setting, there's a high prior probability of disease. I estimate a clinician will often order a test for Marfan syndrome if they think there's like a 75% chance. Yeah, I think this guy has Marfan syndrome. So that would be this orange and beige colored circle. And when you have that high prior probability of disease, the math says the false positive rate is this little sliver over here. Because the false positive rate applies just to the people who don't have the disease. And then the true positives apply to the people who have the disease. So you get this nice big fat positive predictive value. Over here is when you're doing screening. That's when the prior probability of disease plummets. Then the false positive rate applies to the people who don't have the disease, and it starts getting really big. And then the true positive rate is small because the number of people with the disease is small. And so your positive predictive value collapses. And that's the challenge that we face in genomic screening. So we're applying this principle to population screening in secondary findings. And this is a real pedigree of a recent family we've ascertained. And it was a secondary finding, because secondary findings, the math of the screening are the same as population screening. There's a world of differences between population screening and opportunistic screening, secondary findings. I'm not discussing that. I'm just talking about the math. The math is the same. So this man was sequenced for a neurologic disorder, this 37-year-old. And it turns out his mother has a likely pathogenic BRCA2 variant. And she does not have disease. My little annotations here are the age and years followed by whether or not they have the variant, plus, minus, or question mark, and then a letter, no cancer or cancer. So you can do the math here. And using this model, the Bayes model, if you use just the test result on the mom, ignore the phenotype of her or her family, she has a 65% probability of a clinical molecular diagnosis of hereditary breast and ovarian cancer. Now, if given that she is phenotype negative, i.e., she's 69 years old and she does not have cancer, that affects her probability of disease. And in fact, it drops from 65 to 47%, if you consider that. Then you include the fact that she has a 65-year-old sister who is affected and does have the variant. That pushes it way back up. And then if you include all the relatives, people with and without the phenotype and people who either are unknown to have the variant or are known to have the variant, then it goes back up to 83%. So you could see how these probabilities change. Now, you might argue there's not a huge difference between 65 and 83%, but if you look across 48 cases that we've done, here's the distribution of the probabilities of disease in 48 pedigrees. And it's really interesting. It ranges from... I have this little red line here because we determined that the three common variants in the Ashkenazi Jewish population for HBOC are of 100% pathogenicity. So by definition, anyone who harbors that, we say, has HBOC. So they're to the right with a probability of 100. But here's the ones we aren't certain of. And the posterior probabilities range from 22% to 99.98%. And that's an interesting range. And I would argue that the people on the left side of this curve need to be managed differently than the people on the right side of this curve. That's not the same clinical scenario. So I think we need to adopt this probabilistic model and we need to, of course, do a number of things to make this happen. We need robust classifications of variants to determine what ClinGen is doing, the probability of the pathogenicity given the evidence, and the actual numerical value of those probabilities are important in these calculations. Then we need practical methods, and I underline emphasize exclamation mark practical methods to determine the probability of the diagnosis given the phenotype. And I think that's going to have to be patient decision-making support coupled with clinical decision support and define care pathways in the EHR that allow clinicians broadly to be able to do this in a way that supports this probabilistic model and get rid of this foolish and naive concept of determinism, which is this simplistic thinking, if you got the variant, you got the disease, you're going to get the cancer. That's just wrong. So a couple closing thoughts. When I teach this stuff to the Genetic Counseling Training program students, I teach them three rules. I tell them you are absolutely obligated, no excuses if you don't. You have to assess patient risk precisely. I tell them, number two rule is that you actually can't calculate it accurately. We never have all the information. Our information's always incomplete, and some of the information we have is incorrect. It doesn't in any way mitigate your obligation to do it precisely based on the information you do have. And the perfect is the enemy of the good, and so we'll settle for precise and understand that it's not perfectly accurate. And then we have to think about what it means. And what it means for the patient is my third rule, which is that that number that we derive is probably not the primary determinant of care management decisions downstream. That's a much more interesting and nuanced question as to risk perception by patients, and that will drive how they manage their care. And I think so the larger challenge is we have to change our mindset from one of non-directiveness to management. In screening, we are not presenting patients with necessarily the same kinds of questions that we presented them when we did our old-fashioned diagnostic testing. We are not going to be primarily consoling and counseling and helping people adapt to diagnoses. What we're going to have to do is start to motivate people who don't have manifestations of disease because we're going to find them by genomic screening. They're not coming to us with these loaded families. We're going to have to motivate them to engage in these health behaviors in order to mitigate their risk of morbidity and mortality. And we have to do that in a way that doesn't require hours and hours of geneticists and genetic counseling care. It has to be built into the healthcare system more generally so all practitioners can do it. I'll stop there. I think you don't want to take Q&A now, right? Correct. We're going to wait until Mike talks, and then we'll take questions together. Thank you, Les. I'm Mark Williams. I'm a professor at Geisinger, and I'm pleased to introduce my colleague, Mike Murray, who is an internist and medical geneticist. He's currently the clinical director of the Institute for Genomic Health at the ICANN School of Medicine in Mount Sinai. But more relevant to this particular talk, he is co-chairing the American College of Medical Genetics and Genomics Working Group on Population Screening. As you heard from Les, the ACMG has had publications for many years on secondary findings, which we say we should not be using this for making decisions about population screening, which of course led in absence to any guidance on population screening from the college. And so Mike is one of the co-leaders to take that on, and we'll be talking a bit about that. So, Mike. Thanks, Mark. So knowing that I went after Les, I tried really hard to come up with a reverend or a religious leader to put in my title, but I failed. So I'll start with a disclosure, and that is regarding genomic screening. So I've published on where I think we're headed with all this, and this was last year in the publication site at there. But I think eventually we're getting to the point where every individual has comprehensive genetic data set that's generated for their health and meant to be used throughout their lives. So it's going to be linked to an electronic health record, and there'll be two types of evidence-based indications to access it. One is for reiterative screening based on age and other triggers, and the other is for diagnostic assessment. So that's where my head has been for the last 10 years since we started doing genomic screening at Geisinger. Who's ready to do this? Nobody's ready. But there is a group, and I'll mention it at the end, that I think we'll be ready soon to enter into this. How long will it take to get there? I don't think anybody knows that answer, either. So, as Mark mentioned, the American College of Medical Genetics and Genomics has a population screening workgroup that I have the privilege of co-leading with Sonia Erasmussen from Hopkins, and it includes several people in the audience here. And we're taking on the idea of primary findings. So if you create a data set for genomic screening, what screening results should you look for and generate? And so that stands in contrast to secondary findings, as Mark mentioned, that is something that the ACMG has been thinking about and publishing on for over 10 years, starting in 2013 with work that Les led together with Robert Green to think about secondary findings from clinical data sets. The observation was made 10 years plus ago that as we started generating these data sets for diagnostic testing, that you could go back into that data and find things that were potentially useful for that patient and their close relatives, if only they knew about it and perhaps they had no other chance to get that information. So looking into those diagnostic data sets and using it for the secondary use of screening. That was extended to research data sets, including the work that we started at Geisinger in 2015. So using a large research data set that was created for research, but consenting patients to potentially get back that data and use it in the healthcare setting. So this idea of primary findings, what do you do if you create a data set just for this purpose, is something that the ACMG workgroup is tasked with. We've been asked to create two work products. The first is to come up with a rubric or a way of deciding what goes on the screening list. And the second task is to come up with a version 1.0 of primary findings. So a list just like is generated for secondary findings. And we anticipate that that will be versioned over time just as the secondary finding list out of ACMG is versioned. So whenever you're talking about programmatic screening for disease, everybody refers back to Wilson and Junger. This publication is now 55 years old. This is the original cover. This can be found on the Internet. It's definitely worth a look. One of the things that they point out in their principles is that they are screening for disease. And so I want to make a point with the next slide about screening for disease versus screening for risk. And it's a little bit of a different take on some of the stuff that Les just talked about. So pictured here on the left is this bright shiny truck in the 1950s that was a mobile unit put out by the California Department of Health. And so you go in the back door there. You give some health information as an intake. You get a chest x-ray. You get a result. And they were trying to diagnose tuberculosis. We were only about 10 years since the discovery and availability of good antibiotic treatments to cure tuberculosis. So the effort here was to diagnose disease. When we're talking about detecting disease risk, we're talking about something different. And this is of course where DNA and genetics sits, but this is not unique. We do the same thing for blood pressure and cholesterol. So we don't look for high blood pressure or high cholesterol because of the specific need to lower those numbers. We do it because it's risk for major disease, heart attack, stroke, dementia, kidney disease. So when we're detecting disease, the goal is treatment. When we're detecting risk, the goal is prevention and early diagnosis. So in 2021, two years ago, we had a ACMG group that came up with some points to consider around DNA-based screening. That's in your booklet that you received. On the left, we put the 10 principles from Wilson and Younger, 55 years old, but still durable and important. And we targeted them for the genomic screening use. And I'm not going to go through all of them or even read out these four that I've highlighted for you, but these four will be mentioned in the slides ahead. So the first thing that I'd like to mention is this idea that people need to be re-evaluated in some instances. On the left panel at the top there, this is the diagnostic pathway. And for all three of these panels, I'd tell you to watch the yellow ball as it moves across or where it is in space with regards to other things. So in the diagnostic scenario, somebody comes to the health system with a complaint. They have signs, symptoms, family history, something else that drives an evaluation. The evaluation includes genetic testing. When there's a genotype-phenotype correlation, we make a genomic diagnosis or a genetic diagnosis, and treatment goes on from there. When you're talking about screening, the initiating event is the genetic finding. So it's at the start of the line. And then the evaluation goes forth after that. In some instances, that initial evaluation will lead to a diagnosis. There'll be a genotype-phenotype correlation. In others, that patient or that individual should be recommended for reevaluation. And how often and how to reevaluate them will be different for each disease or condition or genetic finding. This is a different take on some of the stuff Les said, too. I think about genotype-phenotype as this Venn diagram. So on the left, you have genotype only. On the right, you have phenotype only. And in the middle, you have this overlap between the two. This is where the genotype and phenotype meet. And this is where, for this context, we're talking about disease or condition that's being screened for. So there will always be in every situation some that have genotype only without disease, including the sickle cell variant and F508, cystic fibrosis. They're smaller in some instances, larger in others, but we have to take this into account when we're thinking about doing programmatic screening of populations. And so this Venn diagram will change, but I think it will always be incompletely overlapping, no matter what the gene disease variant or population. And so some people have heard this in talks that I've given over the last 10 years and said, well, gosh, you know, if you're going to tell people about risk and then they're never going to get to disease, maybe you shouldn't be doing this work. And so I'd remind people, just like Les said, there's nothing new in genomics that isn't elsewhere in medicine. So non-penetrant risk is not limited to DNA. Here's Winnie Langley on her 100th birthday letting her cigarette off her candle. So she was undoubtedly told by many healthcare professionals, probably most of which she outlived, that that was going to kill her, that she was going to get heart disease or cancer, but she proved them wrong. I'd also point out that non-penetrant risk does not necessarily run in families. You all kind of know this intuitively. So identifying a risk in one individual and then identifying it in their mother, father, brother, sister's children, there will be differences in who has disease and who doesn't, and that gives me a chance to mention that cascade testing and the evidence development around that is really important for screening projects at this stage of the game. One of the principles that we talk about is DNA-based screening as a replacement for something that already exists. If we're going to suggest that, it has to be an improvement over the standing way of identifying risk and managing risk. So the U.S. Preventive Service Task Force came forward with recommendations for screening women for BRCA 1 and 2 risk. Here are the three publications over this period of time from 2005 to 2019. And it comes down to this table from the 2019 document. Sorry if you can't read that small print, but it says that the U.S. Preventive Service Task Force recommends that primary care clinicians discuss with women their personal and family history related to the cancers associated with BRCA 1 and 2 and their ancestry. And to decide, essentially, to divide all women into two groups, those on the top row there with enough risk to go forward with discussions and potentially testing, and those on the bottom row where those should not be pursued. Now, when this came out in 2005, I remember having lots of conversations about a lot of people were excited that we had a strategy that didn't cost anything because it cost nothing to get this family history. But if you talk to primary care physicians, they might tell you something different. 10 to 20 minutes of somebody's time to sit down and talk to somebody about this history to record it and then to act on it is something that just isn't operational in any primary care practice that I've ever heard of. I raised this in a number of meetings. If anyone knows an exception to that, I'd love to hear about it because I think we need to know how they're pulling that off. So if people aren't being screened in this way, perhaps DNA is the alternative. We did work at Geisinger Publish now five years ago in this manuscript. Where we took the first 50,000-plus people from the Geisinger cohort and we looked for BRCA 1 and 2 within that cohort. We found 267 cases, which comes out to be about one in 190 people with these variants and this risk. When we went and we said, well, how many of these people or what's going on with these people, we were able to find out that about 20% to be exact, it was 18%. Already knew about that risk or we're already having it managed within their healthcare. What about the other 80%? So that broke up pretty evenly into two other groups. In green there, you see those individuals who had never been tested but also don't meet those criteria, the criteria from the U.S. Preventive Service Task Force and other groups that were actually designed to increase the pre-test probability that if you had a BRCA 1 or 2 test done, that it would be positive. But these individuals, that strategy is not sensitive enough to pick up. In red there are the individuals who did meet the criteria or would have met the criteria but no one had ever interviewed them about it or offered them the testing. So when you ask the question, when you take this strategy to a population, how many people with these variants are unaware but for genomic screening? In this instance, it was 8 out of 10, so 80%. When I first started giving talk about this data, I remember somebody raised their hand and they said, well, you're missing 80%, maybe you're just not doing a good job. Which I thought was probably wrong but I didn't have any evidence till the Healthy Nevada Project out of Reno, Nevada, published their work in 2020 which looked at about 25,000 people. They looked not only at BRCA 1 and 2 but they also looked at Lynch syndrome and familial hypercholesteroemia risk. And they found that about 1 in 75 people have a risk for one of these three genomic conditions and in their data set, about 10% knew about this risk, 90% didn't. So I think we're probably, based on these two studies, looking at 80 to 90% of people finding out about these common genetic risks for the first time through a screening project. The question is how are we going to manage that over time to bring real benefit to those people? So these three conditions are called the CDC Tier 1 commonly and in many publications and that's how I refer to them. So here, if you extrapolate from the data, in 2023, if we did a screen for the nine genes that are associated with these three genetic syndromes, you'd identify north of four million people that have a risk for these, that have these genetic risks. And in the table there, across the bottom are those genetic syndromes but other than the people in this room, not many people care about syndromic names for these things. What they care about is risk for heart disease and stroke, risk for breast, ovarian prostate and pancreatic cancer, risk for colon and uterine cancer. So most people that are thinking about strategies will include these nine genes in their strategies and then what to add after that is certainly a puzzle that lots of us are trying to figure out. I would mention that one of the Wilson & Younger refined criteria that we put out was that clinical implementation strategies should be in place and available to anyone that has the risk identified so that the healthcare system can manage those risks. And when you think about heart disease, breast cancer and colon cancer in particular, we have workflows and systems in place across the U.S. healthcare system to really manage these risks. So much so that I think if you identified those 4.3 million people in the next year or two, the healthcare system would be able to manage them without missing a beat. So we need to think about other risks that we add and how they'll be managed within the current healthcare system. I won't blow the punchlines on David Vintres and other people that are in the audience's work about their March publication from this year, but cost-effectiveness and economic analysis is essential to thinking about screening and we're starting to see good work coming out of that. When you think about how should something get on the list, what does it really come down to? I think what it really comes down to is clinical utility. If you put something on the list and you give back a test result, a screening result, that should prompt an intervention that results in an improved health outcome. Anything short of that, when we start doing this programmatically on large scale or population-wide, I think has to be questioned. Now, over the last 10 years or so, we've got caught up in this term of medical actionability. It's a good term, at least for the initial thinking about what to put on the list. That's the idea that something would have a potential next step if they got a result back. But that's necessary but not sufficient, I think when you start talking about really operationalizing this. And these two concepts obviously are linked together. To paraphrase Gross and Corey, a screening test alone does not have utility. The clinical utility comes in when there's effective access to appropriate interventions. One of the things that I think we should all think about is that we're in this era of genomic screening projects and pilots. And it is really the time for people that are running these projects to think about the engaged population that they have and the engaged expertise that they have and how that can drive a different list for your project or program than what other people are using. So I use the example of the pathogenic founder variance in the A-B-O-L1 gene and the TTR genes that arose in Africa and are very important for populations regarding kidney disease risk and cardiac amyloidosis if there are members of that population that have African ancestry. So not only people that identify themselves obviously as African-Americans, but also those who are Afro-Caribbean and have significant ancestry from other parts of the world. The other thing is that monogenic risk for things like major psychiatric illness, there's now pretty good data that those are identifiable through a screening strategy. Most people that are running screening projects wouldn't have the wherewithal to take that on, but there are some projects and some individuals that are thinking about taking this on and I think taking it on and figuring out if we can find clinical utility by returning those results is a really important task. So again, going back to primary findings from data sets created from four genomic screening is one of the ways that the ACMG group is thinking about it and in my last slide here I'll just cut to the chase and going, I didn't know the rules when I created these slides so thank you Rex for pointing them out, but I'll just cut to the chase down here and I'll say maybe a white paper that I've been thinking about is that a group really needs to put together a timeline and the milestones for completion of the kind of implementation that we're talking about population-wide and then maybe two future conferences since I have the floor that we could think about is, one, we are now in a moment when the projects that are driving population screening, the big projects are mostly privately funded. So when I was referring in the beginning about who's ready to do this, there is money from private businesses and other sources that are ready to start doing this and we have to think about what's going to happen over the next 10 or 15 years. I have no specific complaint about anybody that's done that or that I know is doing it, but when we have a patchwork of the U.S. population, each subgroup of which has a different group of people holding their data de-identified, albeit, but what happens as far as the data environment and the potential uses of that data. I think we have a giant LC problem that's coming. I don't think we should wait until the first real disaster happens. I think we need to start thinking now about the rules, regulations, penalties, these potential criminal penalties for the misuse of data and how we're going to manage that. And the last thing I'll just throw out there is I think we need to start thinking about the Human Genome Project Part II and in my mind what that is is recognizing people having their own genomic data set as a public good for their own health and sequencing the whole population. Sounds a bit crazy, but so did Genome Project 1 when it was first proposed. Stop with my comments and I think Les and I are supposed to sit up here and take questions. Actually, Mike, we made a change. Come back. We're just going to stay here. Awesome. And so what we're going to do, first of all, so thanks to both Les and Mike for getting us started, very stimulating ideas. We expect lots of discussion. We have 45 minutes for discussion, which is great. And thanks for staying on time. We're going to alternate. So I can see this side of the table, but I can't see this side. But Mark can see this side of the table. So we're going to alternate. And what would be helpful is, yeah, if you do these sorts of things when you have a question, then we will keep our eye out and we will call on people. That includes people who are not at the, who are sitting in the second row. Please, you know, we should be able to see that as well. So I can't see the name, but go ahead. I'm Kate Nathanson from the University of Pennsylvania. So I had a comment about Les's sort of thinking about this. I personally, unless you didn't mention, I'm struggling much more with the issue of somatic testing on tumors and then identifying variants that have, and reflex testing, particularly, and I don't know how many people follow us at all, with the new ESMO guidelines, which are extremely broad in terms of reflex testing. And one of the things I'm personally struggling with and too bad, Bruce, is in here, is NF1 reflex testing in the context of people without a clinical diagnosis of NF1. So I just want to bring that up as another form of population screening that we should be aware of and considering when you're talking, when you think about this sort of secondary findings. This is another whole set of secondary findings that I think have a big impact. And I'll just make, Micah, a quick comment that we've done obviously the same study, although not published, looking at our rate of B.R.C.1 at two mutations and how many no and 10. It's much higher, actually. It's about 50%. So. Yeah, I guess two responses that, first is that, of course, if they're having somatic testing, there's a reason why someone did that. So that elevates their prior probability of disease. So that's really not population screening. That's working up a patient with a disease. Given the list of genes now that they're doing, I mean, that's probably... Oh, so for genes other than related to their disease and somatic tests, I get you. It's essentially secondary findings. Got it. So the list is probably 30 genes maybe longer now. So that is... I understand what you're saying, but no, this is essentially like you do someone who has, I'm going to say renal cancer, and you essentially find a B.R.C.1 mutation. It's not an associated cancer. You incidentally find it, and it is population screening. I'm just saying that's a whole other thing you're struggling with where you're... How do you manage that? Yeah, thank you for the clarification. So there's two issues there. The number one is that really, what you're talking about there is lowering the reliability of the analytic validity of the finding, right? Because that finding that somatically is not as good as finding it constitutionally. So that's number one. And then number two is an interesting debate going on. I'm not a cancer epidemiologist, but my colleagues who are tell me there are some questions regarding whether or not all genes that are cancer susceptibility genes are pan-cancer susceptibility genes with just different relative risks and that our numbers aren't adequate to demonstrate that, but they think there's a very long tail with a risk, and that's a different issue. I'm happy to have a big debate about that. But this is, in the context of the recommendation, you said somatic, so there are identified somatically, and then people are getting reflex testing for them. So they're being identified as having germ limitations through the reflex testing. So there's a really broad set of genes now that are suggested where you should have reflex testing. So you are identifying people who have unrelated diseases incidentally through reflex testing. Sorry, just to clarify, that's happening. And so to me, that is another form of population-based screening, to be honest. Or maybe perhaps it's somewhere in between a population screening and opportunistic screening. Yeah. Yeah. Great. I'm going to take moderator's prerogative and ask a question, also to Les. I love the concept of the practical probabilistic model of population screening. And it struck me that we could potentially get to something that looked like the Richards criteria for variant interpretation, where you say, here are the different pieces of information that you need. Here's how you assess them in a rigorous way so that people could move away from doing them as one option, just kind of say, here's the information we need to find, and here's the information we need to generate to be able to do that. Is that the direction you think we need to proceed? Absolutely. And we're collaborating with, I don't know how many of you know this group at Harvard, Giovanni Parmigiani, who has this website called Ask To Me, which is called All Syndrome's Known to Man. And the gender specificity of that. And it's basically epidemiologic meta-analyses of cancer risk for all these syndromes by age. And then they back-calculate the likelihoods of disease, the likelihoods of finding a variant in a person who has disease. So all the data are there. And then, again, per the reverend, you just flip that into the inverse probability, and you can ask the question, if you have disease, what's the phenotype, what's probably have the variant in such. So we can do this. And he has a very simple plug-in tool, and we're just going to have him flip that and then provide, oh, I have this person with this variant, and they had cancer when they were 63. What's the likelihood? Or their ant has cancer? And we can derive, because the cascade testing that Mike brought up is as essential as it is difficult to do practically. But we can begin to derive risks from even family members who aren't genotype using Mendelian principles. And so you can do simple plug-in, computer-based algorithms that will spit out the risk of disease based on that input data. And you shouldn't have to be a Bayesian expert or a geneticist to be able to do that. I think I saw Carol's hand up. I think Rex was first, and then Heidi. Well, he took his pick, and that's why. So it was my turn, and then we immediately found a flaw in our system, because I'm covering this and somebody in the middle. So I'm going to start with Carol, but I will use Rex next on my next pick. Carol. Very good. So both talks brought up a very similar question in my mind that might follow on a little bit from Mark's comments and questions, and that's evidence. So the probability of pathogenicity given evidence and the amount of evidence we need to move something into a tier one. And so what are our gaps there? What are the areas that we can improve the evidence or the process to determine pathogenicity and what are the types of evidence we need to move something into something like a tier one screen? This is a lot of that question. And I think it's one of the problems I have with the tier one. And I think grouping together, for example, HBOC with FH, I think was crazy. And the numbers are one thing, but I think what we have to do is we do have to understand the numbers and we have to understand them precisely how predictive power is and how likely we are to change the health care outcome. But we also have to have a really sober discussion of our error modes. What happens when we have a false positive? What happens when we have a false negative? And I would argue that it's a completely different situation if you have a false positive for HBOC as to the adverse health consequences of that versus FH. I couldn't care less about a false positive for FH. It's not going to ruin anyone's life, guaranteed. I would think very differently about HBOC. And so we've got to understand the numbers and we have to have our community has to get together and make really clear headed decisions about we're going to do this, here's what happens if we go down this pathway and here's what happens if we go down that pathway. And so it's not only the numbers, but it's a values discussion about what errors we can really tolerate. So I mentioned that one of the evidence gaps, one of the evidence gaps that we're going to face is right now we have to extrapolate outcomes from data that's derived by a different methodology. So we say that BRCA variants are 80% to 90% associated with disease over lifetime. That's based on different data than we're generating through screening. We don't know that number for screening, but we need to get to it. And places like Geisinger now have almost 10 years of experience. We probably need 20 or 30 to really get those numbers, but we got to get to them. It's going to be lower and it's going to be different. Great. Caitlin. Thank you. Maybe we should name sides of the room, right? So team, the team have teams down the middle. I'm Caitlin Allen and I'm at the Medical University of South Carolina in Charleston. So Les, you mentioned motivation and I think Mike, you mentioned utility, buy-in. And so I'm thinking about this from more of an implementation perspective. And maybe adding into that equation, we've talked a lot about the clinical utility, but thinking about research utility. And so for our population screening program at MUSC, that's been really a huge motivating factor for us is that we have now a large population of genomes that we can then go back and do research. And that's been huge to get patient buy-in for participating and also our provider buy-in and leadership buy-in. So just maybe adding that to the equation or curious to hear any thoughts about sort of the research component of motivation and buy-in. So I think in every setting, we should be getting participants to be offered the chance for their data to be used as research. Yes. So asking every participant that gets screened to be involved in research is essential. I think if we're going to do this right over time, we need as many people participating as possible. So that motivation, which most people will agree to, is really important. So I think what you're doing there makes a lot of sense. Yeah, with just that, I think it overlaps on genome medicine 14, which is this silly notion of this huge wall that divides research and clinical practice. We just have to get over that. And every research participant should be seeing the clinical benefit of their participation, and every patient's data should contribute to every other person's health care through the research mechanism some way. We just have to make that happen and think more openly about that. I can just clarify that Genomic Medicine 14 was on Genomic Learning Health Systems, which do exactly that. That basically every patient is a research participant. Every research participant is a patient. I just wanted to make a clarifying comment. Mike, there obviously have been several popularly very well done case control studies looking at risk in BERSA1 and 2 as well as other cancer susceptibility genes, which I think is what you would want to be looking at, both carriers and bridges, which were published in New England Journal in 2021, and a lot of resulting studies. And the penetrance for BERSA1 and 2 done those well done epidemiologically matched case control studies was about 50%. So there really, those data are there. And I think that they were done in studies that were really well known like nurses health and things like that. And I'm sorry that Pete's not here also to comment on that. Thanks. Okay. I would also just note that Mike's point is still relevant because the case control study still involves somebody that has phenotype. And in population screening, we're talking about ascertainment, irrespective of phenotypes. So there's still a gap there. Heidi Rehm. Thanks. So as we get better evidence, and we better understand probabilities, we can better feed into, you know, recommendations for clinical screening and follow-up. And we have a great framework and flingent for actionability and all the variables. But I think one aspect of this that I anticipate being a challenge is that patients' perception of what is important to them or not differs. And I think a great example of that is amniocentesis and the risk of losing a pregnancy versus the risk of a child with a chromosomal abnormality and where those lines intersected became the age a woman shouldn't get an amnio. And before that, they shouldn't, as if the risk of loss of a pregnancy was equivalent to the risk of a child being born with a chromosomal abnormality. And obviously for each individual woman and couple, those may be very different concerns. A couple that took 10 years to get pregnant and there's one that took one day to get pregnant. So as we think about getting good at the front end of this, do we just leave it to the physician to have to have the long discussions about patient preferences or how are we gonna build in sort of scalable approaches to patient preference is my question. That's a great question. That's probably for genetic counselors in the room to respond to. A session on that. Good. And I think what we need is to understand the basis, the fundamental basis of what these counseling decisions are and figure out a way to make what is currently a 45 or 60 minute patient encounter, which is totally unaffordable, totally not scalable. And if it's required is the end of genomic medicine. Don't wanna overstate that. But I did anyway. But if we could understand how this process works, I would suggest in the large majority of cases we could build computer based online interactive tools that could help people understand where their values and their key decisions that they want to, or outcomes that they want to maximize or minimize and say, look, a person who thinks about this the way you do generally goes down this pathway for these reasons. And I think we could supplant a large majority of this decision making from this super expensive one-on-one patient encounter to tools to do exactly what you're hoping. I think I saw Jonathan's hand up. Okay. I'm sorry? Oh, wait, that's that side. Dude, I may be on this, I may be on this, Jeff. So, Jeff and then Jonathan, Dex. Go ahead, Jeff. Thanks. One of these lot of referees around here, don't we? Tough crowd. They... Jeff. This is Jeff Roscoe, and actually it works well because mine's a good follow-up to Heidi. So, I'm a public health person at HRSA, and you mentioned about newborn screening might fit in certain as learning lessons. I was really struck, Mike, by the picture of the woman with a cigarette at 100 years of age and how hard it is right now to get health behaviors to change. And particularly your comments about clinical utility. So, thinking about newborn screening, sickle-cell disease, we have great ascertainment. We've known the gene for decades. We're getting better at some of the genotype-phenotype correlations, and yet still fewer than half of children get hydroxyurea transcranial doppler and it's even lower in adults. And there's such a huge health systems issue right now with things that we are very clear about. It's one of the biggest challenges I think we face. Not that there's the answer to that, but just to bear all of us to remember that we may have systems in place, but man, they're not really getting to everyone. I just follow up on that. It's a really important comment, and again, that's part of our changing our mindset from dealing with families who come to us highly motivated already with dealing with the John population, which is mostly unmotivated, or I don't know, is it negatively motivated? What would you call that? I can tell you in our secondary findings, study of families that we're ascertaining includes FH, exactly zero of the children in those families have been screened for hypercholesterolemia, and that's an AAP recommendation. Routine care, and zero. We encountered a single one. It has no penetration. So you have to make this as automated and simple as we possibly can. And I just add that, you know, this is one of the reasons why actionability and clinical utility don't equal each other, and we've got to understand the reasons for that and come up with better systems for getting people from an actionable result to a clinical outcome that meant. Well, you have an optimistic note though, and at least the studies that have been publishing their data to this point, more than half of individuals that have been received a result have actually made a health behavior change of some type in terms of at least getting a test ordered or something of that nature. So it is encouraging that it seems like this may be having some effect more so than what we do typically with diet and exercise and those sorts of things. Ned. Thanks. I really enjoyed the two talks, especially being right next to each other, trying to navigate the information between the two is fascinating for me. So, Michael, you tell us there's a lot of undiagnosed variants out there unless you tell us not all variants are created equally or uncreate, I don't know how to say, have the same risk. Some are more risky than others. And I think about applying that in the setting of population based screening where you're going to generate a lot more tests. That's what Jeff was talking about, the impact on the system overall. And I think about clinical utility and what I didn't hear again was the heart because if you're going to generate more uncertainty, you're going to generate more positives with variation and penetrance and we're going to leave it up to an unprepared medical workforce to make decisions. We are going to buy more mistakes, more errors and more harm. So as we think about clinical utility, always think about not just the good things, right? But all of the harms that we're going to accrue because of population based screening and the hope, right, that we all hope is that when we balance it all out, we're doing more good harm. But I just ask the group never to forget the kind of downside to the results. Great points in both. Related to that, when we talk about false positives, I was thinking as this is being discussed, we need to make sure that we're distinguishing that from non-penetrance. So some people just blend those two together. And so biologic evidence that a variant can cause disease is pathogenic, is different than giving back a variant that the data was misinterpreted and is actually not pathogenic. And that will hopefully help to clarify some of those issues. Yeah, I would just add to that. There's a wonderful quote I came across when preparing the talk. Isaac Asimov said, the uncertainty that comes from knowledge is not the same as the uncertainty that comes from ignorance. And it's really important to remember that. And we in medicine are extremely good at washing our hands of awful things that happen because of our ignorance. And our only choice here is to make decisions based on no genomic information or based on our decisions based on genomic knowledge. And we can make better decisions with more knowledge. I don't think we should ever confuse that. Not to say that we won't make mistakes. And per your question earlier, we have to do error mode analysis and say how many people are going to end up in this error mode and how many people are going to end up in that error mode and we make rational decisions. We grossly, newborn screening is a great example, right? It has a very high false positive brain. And the whole system is built to work with that high false positive rate because we insist on the sensitivity being high because avoiding a kid having a devastating disorder like PKU is the outcome we really want to avoid. And so we way over diagnose it and then we get rid of that in the follow-up testing. Not to say we should do that in genomics, but we have to say, what do we want for the false positive rate? What do we want for the false negative rate and where is the right trade-off? And I would say that those trade-off numbers are completely different for FH than they are for HBOC. And we have to have that discussion and say, okay, where should we set those? Because we don't get perfection and perfect is the enemy of the good here, but what's the optimal good we can get out of this? One more just basic orientation piece. I think that these types of questions have a very different context if this meeting was about, are we going to implement population screening or not? This meeting is about what are the research questions that need to be answered if we're going to proceed to population screening. So these types of issues where we may have fundamental differences, mostly those differences are being driven by, we don't have the information, which means they're really a high priority for research. And so for those of us that have been tasked to capture information, which gives me another opportunity to be criticized about how I'm doing here, we will be contextualizing these types of debate points into a research agenda. Jonathan Burke. Jonathan Burke from UNC Chapel Hill. The first two talks are why I came here. This is so good. I'm also giving a talk, but this is thanks for the setup. One sort of commentary thinking about the two talks is given the evidentiary requirements for doing this and getting anywhere close to right. It's going to be perfect. I think it really drives us to the conclusion of starting with a very, very small and very well understood set of conditions. Just so for the ACMG committee to think about, let's not go to 70 right away. Let's try to focus it in. And I guess the other sort of point about that is if we're not really close to that sort of whole genome analysis and healthy people goal, as you kind of pointed out, what would you think about at least as a starting point for that considering whole genome analysis in consenting adults before we go all the way to whole genome analysis in all new boards? That seems to me like we could get a long way towards the goals of identifying all those risk factors and getting our feet fully wet before we start into new boards. So we have this interesting issue that will come up. If we do a whole genome and then we limit our analysis to the CDC tier one, won't somebody take that data and say, let's go for secondary findings here and look for all other kinds of things? So then do you recommend doing just a limited data production to only look at those things? Well, then you're talking about wasted opportunities from a budgetary standpoint. So it's a conundrum. The one thing that I didn't mention at the front, which is worth mentioning is the ACMG has been very clear over the years that the secondary findings list was not generated as a population screen list even though everyone has assumed that from the beginning. But when you think about it, the secondary findings are from clinical data is generated within a patient provider context where that provider, by ordering that test, takes on the responsibility to see through the outcomes related to that test. So it's going to be very different when we're doing screening of populations and no one's really in charge of kind of seeing through all those variants. And we've got to address that too. George. Mark, thanks. Actually, since I put up my card, my questions have evolved. And I'm not sure I remember the original question I wanted to ask, but I love that quotation about uncertainty and knowledge and ignorance. And I'll be honest with you, increasingly, the challenge is distinguishing knowledge from ignorance. A lot of people tell us they've done their research and you dive into it. It gets a little difficult. So let me make it brief and thank Mark for reminding us of the core purpose for what this session is about and maybe make a friendly amendment that as we address the research directions for when to screen, who to screen, how to screen, maybe also think of the engaging the people we screen and the research questions about implementation. I think they're really very relevant if we could extend the discussions to that. And then finally, there's something about the automation and use and artificial intelligence to extend how much we can do. All for it. It's really terrific. The challenge we also beginning to see is that there's a segment of the population that have no access to that. And so as we do more and more of that, we tend to exaggerate the disparities that already exist. And so what are some thoughts around how we overcome that before everyone has full access to automation and AI and machine learning? Yeah, I would say I'm really a proponent of both opportunistic and population screening because I think as it's currently structured, I can't imagine a healthcare resource that is more disparate in its availability to patients than genomics and genetics. I mean, the obstacles that people have to go through to get to see a person like me or Mark are ridiculous and only the most resourced, educated, advantaged people can navigate that gauntlet. It's ridiculous. It's terrible. And that's disparity. And by making it as broadly available as we can, we have an opportunity. Not to say that it will eliminate disparities. It won't. Not to say it will make irrelevant all of the other disparities in our healthcare system. It won't. But it will begin to level that playing field of access to genetic and genomic healthcare. And I think we have to push really hard for that. We have a lot to go through, but we're doing well. Christine. Sure. Thank you. So as we, you know, want to continue to improve our understanding of our communication of risk and penetrance and probability, it seems that, you know, the screening is not going to be just a one and done type of deal with the screenings that we're going to have to continue to engage them so that we understand their outcomes over time. And I was interested in what Mark was saying about how I think you said 50% of screenings are taking at least one action right after the results. But how do we, or what are the thoughts about ensuring that we continue to engage so that we have a fuller understanding of outcomes? I think the easiest way to answer that is to stay tuned because we have sessions throughout the meeting that are going to be really focused on that. So if you're willing to kind of put your question in the parking lot, hopefully we'll get some answers and some interesting research opportunities for that. Is that, that's not an answer, but it's practical, I think. I'm going to go to Terry because there's an online question from April Adams. Because you put, well, yes. I mean, because I'm not in the meeting. So yeah, go ahead. Go ahead. It may not be the onus on that individual. It may be that our health care system is prioritizing people who have resources, people who have high health literacy, you know, and not prioritizing people. So I think starting in a room like this, with a lot of stakeholders and having language that is not putting all of that on that individual is a really big place to start when that brings you down. Why does an equity matter? Why does a labor city matter? How are we going to actually, does it make them yield more of ours? And we'll probably talk about that. Yeah, I would just add, we do have, I think, an outstanding session this afternoon that will address at least some of those issues. So stay tuned. I know there was two down here. So here, props. Yeah. Let you go. Thanks, I'm Kelly. He's from Hudson Alpha. This kind of circles back to, I think, Heidi's comments earlier about patients and bringing their kind of preferences into ultimately what turns into outcomes that we can't necessarily script that, that there's all the other variables at play. And I guess just wanted to bring up the additional variable of just patients believing the results that they get back and when it does or doesn't corroborate with their story, their narrative, their history. And we bumped into people that, you know, get a negative result and they flat out say, well, I don't, I don't believe it. You know, I think that's wrong. They get a positive and they say, well, I don't believe it. And that's going to play into those actions that are taken. And, you know, in those conversations, sometimes we can change their minds. Sometimes I'm convinced we don't change their minds on whether they believe it at the end of the day. But as we scale things up, you know, just making sure that we're kind of looking for ways to, to improve that belief and have those narratives discussed and approach that with populations in more scalable ways that I don't know the answer to that, but that's going to play into the ultimate impact and benefit of patients. Thanks for that comment. Dan Roden. Yeah, I have two probably irrelevant questions. One was, one was from Mike. You mentioned at the end this idea of monogenic psychiatric disease. Maybe you could amplify on that because I'm not sure I know what monogenic psychiatric disease is. There's a whole polygenic psychiatric disease. Problem that, you know, people are using polygenic risk scores to find people with schizophrenia or depression. And that's going to be an interesting implementation issue. That was, that was sort of an information question for me and maybe I'm the only one in the room that doesn't know what you're talking about. But, and then the other question was for, for less than it may be that we'll deal with this later. But the, if you're, I love the Bayes theorem sort of approach to this that makes perfect sense. The, if you have a patient who has a phenotype and you're then going to, and the phenotype could be cancer susceptibility, that could be a phenotype. Or the phenotype could be something that you can actually put your hand on like cardiomyopathy or something like that. How, maybe this is a question for actually Jonathan Berg too. How far into the long, long, long, long list of potentially causative or associated genes does Bayes tell us to go? So, you know, if I have somebody who has a cardiomyopathy there's four genes that I might want to look at that have high, that have 10 genes or something like that. But then there's another 150 that somebody wants me to look at. And how do I, how do I sort of balance that? How do I approach that? And I'll tell you why at the break why I'm asking this particular question. But I'd be interested in your thoughts. So the psychiatric, monogenic psychiatric disease I'm just not. Yeah, so I will send that reference around. I've been approached by people, I Yale and now at Sinai that are interested in looking into that. But it's a good point. What gene? I'm sorry. And I think we were also, you know, focused on single nucleotide variant here. But the real reality, and this gets into the polygenic if you will, is that we can do copy number variant screening and there's publications out of Geisinger that have shown that we are not doing diagnostic testing of a lot of people that carry known pathogenic copy number variants like 22Q11.2. And the vast majority of them, if not 100% of them, have neuropsychiatric complications related to that. And so if you want to constitute that as a monogenic screen disease, I think there is some evidence of that. OK. You're in a second question. It's awesome. And in fact, we're incorporating that concept in an upcoming paper and it will be in the version four of the ACMG AMP recommendations, which is that negative evidence, if you have a phenotype that has locus heterogeneity, negative evidence at one locus comprises positive evidence at other loci. That was not in version three of Richard's et al. But that's what you're saying, right? Because when you start once, was it, who was it? Was it Sherlock Holmes? Once you've eliminated? It would be impossible. Whatever remains however improbable. Right, right. Absolutely. That's it. That's basic. President Conan Doyle actually, who said that? Yeah. And so I, you know, that's why I'm a genome fan because I want all the variants. And then I want to do a probabilistic assessment starting with the most likely culprits. And once I've excluded them, I'm walking down that curve of contribution until I get to the most likely answer. Does that make sense? Sort of. It doesn't tell you when to stop. Because once you have the bird in the hand, the question is do you keep looking in other bushes? Right. That's an interesting question. Bays won't tell you whether or not to do that. I'll just tell you what the likelihood is that you might find something else if you keep walking. Do I start with the 10 genes or do I start with the 200 genes? I think you should always have all 200, but you should start with the 10. I like that. Jillian. Hey, Jillian Hooker, our concert genetics. I want to go back to the conversation that Heidi started, Les's response, and some of Kelly's comments about genetic counseling and patient preferences, and even the scalability of genetic counseling. And first point out that from a numbers perspective, genetic counseling is one of the most rapidly growing professions, and the numbers are a lot bigger today. Then they were five or six years ago when we were talking about a shortage, and I think many folks have sort of acknowledged or sort of accepted a shortage as a truth without actually like digging into what's behind that. And in fact, I think many program directors right now are worried that perhaps they're training too many people, given the contraction in the lab market that they're increasingly worried about the job market for genetic counselors right now. But I think what we have then is not so much a shortage, but a distribution problem of where genetic counselors are, where positions are, which is related very much to the finances of it, which is sort of covering these positions, which are also on their way to getting a little bit better with a new CPT code slated to come out in 2025 that I think will really, really help in that situation. And then to the like we can't scale to individual level interactions for all folks who are going on to engage in screening, I think already we're doing that in MRI and echocardiogram colonoscopy. These are individual provider patient interactions, most of which cost far more than a genetic counseling appointment, right? And so I think there are economic models that might be more sustainable for follow-up to population level screening where genetic counselors could actually save costs in a way that would make them more scalable. Unless I'm missing something, which I guess is my question, this is the piece I'm not seeing there. Labor will always be the most expensive part of healthcare. And anything that we can do in healthcare that supplants an hour of any provider's time with some technique, some technology, some non-human interaction that can do most or all of what that human does will save money. And that you're right that genetic counselors per hour are less than geneticists, but they're still, that's still real money. And what do you think an hour of genetic counseling actually cost the healthcare system in total? It's probably $200 or $300? We'll have the answer from the rock. We haven't done that yet. There's a study going out right now. That's actually being calculated as we speak. So there will be at least a Medicare-related answer to that question. Yeah. And it will most certainly be less than the cost of an MRI, an echocardiogram, or a colonoscopy. Yeah, I think that no one would debate that at all. The other editorial comment I just make is that numbers are one thing, distribution is another. And so the distribution of the resource is also something that's important to consider. But I think this will be a great research question, which is how do we actually efficiently deliver this if in fact we move it forward? We're going to go back online to Carol Horowitz. Hi, everyone. I'm so sorry I can't be there in person. Follow up on this conversation. I think it's not just from a genetic counselor and the distribution. I think more research is needed into when we need genetic counselors. I was struck when we have now tested many thousands of folks for equal environments that almost none asked for and had any genetic counselor, even though it was offered to all of them for free. And I think we need to think about when this is, as Terebinolia you mentioned many years ago to me, is a goal of like a creatinine test. Do we need a genetic counselor to go through that? And how do we start looking at all the different tests we do and what the resources are needed to give good quality equitable care that's accessible to many people? We have Bruce Korf online who has a question. That's going to be the last question. Bruce? Yeah. I'm one of those that had an unexpected encounter with an unknown small genome. I'm sorry. I want right now. Actually I was going to make a comment. I've just been delayed. I think Kate made the point about finding NF1 variants in people who had cancer testing, usually breast cancer testing. And we've seen a fair number of people that don't have any symptoms. Occasionally they just didn't understand what they were. Or some of them maybe was AX. But some actually really don't have anything that you can find in most of the world. I think it may turn out to be the case. We haven't proved this yet. But some of them may have a totally bad oasis. And we're also doing an analysis of all of us today now and we find a pretty substantial number of known pathogenic NF1 variants where there's no tracing record of any clinical diagnosis. And I think I'm not sure they all have that. No doubt some of those. But I suspect some too may have totally bad oasis. No, though maybe a small point. It's something just in mind as we look at older populations. That's what we end up doing with screening. One other point, which is a little bit, which is based on comment just made a woman go about genetic counseling. You know, I take the point about the distribution, but I would also make the suggestion that we will never, ever have enough genetic counselors or medical geneticists to see the people who are involved in screening. And that we need to think in different terms in looking at ways to communicate information. I personally think that the genetic counselor of the future will include artificial intelligence in the space sort of the chance that people will be able to converse with and get information as these systems are approved, but I don't think we'll be able to do that. Any comments from the speakers? Last comments you want to make? Less is good, Mike, you're good. Okay, well we're, first of all, let me thank both speakers for stimulating our thinking in this opening session. Lots of great questions from participants. We did leave some tent cards up. We recognize there's more interest than we have time for, but this is just the first session and I'm sure you'll work your questions but we will take a break now and we will reconvene at 11.05 for session number two. Thank you.